Few-Shot Object Detection via Variational Feature Aggregation
Authors: Jiaming Han, Yuqiang Ren, Jian Ding, Ke Yan, Gui-Song Xia
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on PASCAL VOC and COCO demonstrate that our method significantly outperforms a strong baseline (up to 16%) and previous state-of-the-art methods (4% in average). |
| Researcher Affiliation | Collaboration | 1NERCMS, School of Computer Science, Wuhan University 2State Key Lab. LIESMARS, Wuhan University 3You Tu Lab, Tencent {hanjiaming, jian.ding, guisong.xia}@whu.edu.cn, {condiren, kerwinyan}@tencent.com |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions implementing their method with Mmdetection but does not provide any concrete access information (e.g., specific repository link, explicit code release statement) for their own source code. |
| Open Datasets | Yes | We evaluate our method on PASCAL VOC (Everingham et al. 2010) and COCO (Lin et al. 2014), following previous works (Kang et al. 2019; Wang et al. 2020). |
| Dataset Splits | Yes | We use the data splits and annotations provided by TFA (Wang et al. 2020) for a fair comparison. For PASCAL VOC, we split 20 classes into three groups, where each group contains 15 base classes and 5 novel classes. For each novel set, we have K={1, 2, 3, 5, 10} shots settings. For COCO, we set 60 categories disjoint with PASCAL VOC as base classes and the remaining 20 as novel classes. We have K={10, 30} shots settings. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper states 'We implement our method with Mmdetection (Chen et al. 2019)' but does not provide specific version numbers for Mmdetection or any other ancillary software components needed for replication. |
| Experiment Setup | Yes | We adopt SGD as the optimizer with batch size 32, learning rate 0.02, momentum 0.9 and weight decay 1e4. The learning rate is changed to 0.001 in the few-shot finetuning stage. We fine-tune the model with {400, 800, 1200, 1600, 2000} iterations for K={1, 2, 3, 5, 10} shots in PASCAL VOC, and {10000, 20000} iterations for K={10, 30} shots in COCO. We keep other hyper-parameters the same as Meta R-CNN (Yan et al. 2019) if not specified. |